mortalidad infantil e inequidades socioeconómicas brasil

8/12/2019 Mortalidad Infantil e Inequidades Socioeconmicas Brasil

1/23

Trends in Socioeconomic Inequalities in Mortality in Developing Countries: The

Case of Child Survival in So Paulo, Brazil

Narayan Sastry

Demography, Vol. 41, No. 3. (Aug., 2004), pp. 443-464.

Stable URL:

http://links.jstor.org/sici?sici=0070-3370%28200408%2941%3A3%3C443%3ATISIIM%3E2.0.CO%3B2-0

Demographyis currently published by Population Association of America.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtainedprior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content inthe JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/journals/paa.html.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. Formore information regarding JSTOR, please contact [email protected].

http://www.jstor.orgFri Apr 6 10:45:45 2007
http://links.jstor.org/sici?sici=0070-3370%28200408%2941%3A3%3C443%3ATISIIM%3E2.0.CO%3B2-0http://www.jstor.org/about/terms.htmlhttp://www.jstor.org/journals/paa.htmlhttp://www.jstor.org/journals/paa.htmlhttp://www.jstor.org/about/terms.htmlhttp://links.jstor.org/sici?sici=0070-3370%28200408%2941%3A3%3C443%3ATISIIM%3E2.0.CO%3B2-0


2/23

TRENDS IN SOCIOECONOMIC INEQUALITIES INMORTALITY IN DEVELOPING COUNTRIES: THE CASEOF CHILD SURVIVAL IN SAO PAULO, BRAZIL

N R Y N S STRY

I exam ined trends in socioeconomic inequalities in und er-five mortality for the state of SGoPaulo, Brazil , o ver a 21-year period from 1970 to 1991, during which mu ch of the mortali tytransition unfolded. During this time, there was a decline in inequality in under-five mortali~ yhousehold wealth but a substantial increase by mother education. Improvements in infrastructureand econ omic develop ment were associated w ith lower levels of socioeconomic inequali ty inunder-five mortality. Mother education emerged as the key factor underlying socioeconomic in-equalities in under-five mortality even as levels of education for women increased and inequality inschooling fell.

In the past few years, there has been growing policy and research interest in socioeco-nomic inequalities in health and the health of the poor in less-developed countries(Gwatkin 2000). Reducing health inequalities by socioeconomic status and improvingthe health of the disadvantaged have become central goals of the World Bank, the WorldHealth Organization, other international organizations, and major donors of developmentassistance (Wagstaff 2000). Yet progress toward achieving these goals may have beenstymied by an important gap in documenting and understanding trends in socioeconomicinequalities in infant and child mo rtality in less-developed c ountries.* Few studies havedescribed trends in these measures, and even few er have sou ght to explain them. Little isknown about how socioeconomic inequalities in health have changed over time as thedevelopm ent processes unfolded and levels of urbanization rose, wom en s educationalattainment improved, infrastructure spread, and income and wealth i n ~ r e a s e d . ~raveman

*Narayan Sastry, RAND Corporation, 1700 Main Street, P.O. Box 2138, Santa Monica, CA 90407-2138;E-mail: [email protected] . I gratefully ack now ledge the research assistance of Gab riela Lopez and Stepha nieWilliamson and financial support from the National Institutes ofH eal th through Gran ts HD 3669 5 and HD38556.1. Socio econo mic inequalities in health refer to the exten t to which difference s in health status are system-atically related to differences in socioeconom ic status.2. Num erous s tudies have examined the effects of socioeconomic status on health or mortality using cro ss-sectional data. However, few of them have extended their findings to characterize levels of inequality usingeither rate ratios or, especially, more sophisticated m easure s of inequality (described la ter). Additional com pli-cation s of extractin g information on trends in socioe conom ic inequalities in health from cross-s ectiona l studiesare that the specific measures o f socioecon om ic status often differ across studies, as do the number and type ofother variables that are held constant.3. Cleland , Bicego, and Fegan ( 1992 ) found that disparities in child survival by socioeconomic status andmaternal e ducat ion did not narrow from the 1970s to the 1980s in a dozen develop ing countries . Wagstaff s(2002a) reanalysis of the results from a number of studies showed that inequality in under-five mortality in-creased in Bolivia from 1994 to 1998, in Vietnam fro m 1993 to 1998 (Nguyen and Wagstaff 2002), and inUganda from 1988 to 1995 (Stecklov, Bomm ier, and Boerma 1999). In addition , Victora et al. (2000) reportedno chan ges in inequality in child ren s health and infant mortality in two are as of Brazil from the late 1980 s tothe early 1990s. Several studies have shed som e light on this issue by exam ining differences in covariate effectsover time. Fo r example, DaVanzo and H abicht (1986) fo und that the effects of mother s education on childsurviv al increased ov er time in Malaysia , and Merrick (1985) found that the effects on children s survival ofmo thers education decreased but those of fathers education increased in urban Brazil between 1970 and 1976.Demography Volume 41-Number 3 August 2004: 443-464 443
http:///reader/full/[email protected]:///reader/full/[email protected]


3/23

444 Demography Volume 41-Number 3 August 2004and Tarimo (2002: 1622) noted that relatively little informa tion is routinely available onhealth status or health care disparities between better- and worse-off groups within mostcountries, and particularly on how within-country social disparities may change overtime. In contrast, much research has documen ted trends in poverty and income inequal-ity ove r time and has analyze d the factors underlying these trends-both in Braz il and inother parts of the developing world (see, e.g., Lam and Levison 1991; Psacharop oulos etal. 1995).Research on socioeconomic inequalities in health in less-developed countries has,until recently, focused on disparities based on categorical indicators of status, such asregion, rural-urban place of residence, mother's education, sex, and race or ethnicity. InBrazil, attention has focused primarily on rural-urban and regional differences in mortal-ity, particularly between the poor and underdeveloped Northeast region a nd the rest of thecountry, and secondarily on difference s by income and education (Sastry 1996).4 The tre-mend ous geographic diversity in the level and pace of social and economic developmentin Brazil has been the principal reason for this focus; differences by socioecono mic statushave tended to mirror the regional differences and hence have received less attention.Another reason is that many past studies used basic indirect estimation techniques. Be-cause stratification of the sample was necessary for estimating mortality differences, thesestudies were restricted to exam ining the effects of one or two cova riates at most.Less attention has been paid to ex amining the effects of broader social and economictrends on declines in infant and child mortality and on changes in disparities inmortality. One exception is a study by Victora et al. (2000), which examined trends inthe inequality of children's health by economic status in Brazil. These researchers foundthat in Cea ra state in Northeast Brazil, disparities in children's health and infant mo rtal-ity between the rich and the poor remained largely unchanged between 1987 and 1994.For the city of Pelotas in southern Brazil, the situation was largely the same from 1982to 1992. However, when the analysis of inequality in infant mortality for Pelotas wasstratified by low birth-weight status (there were no other controls), Victora et al. foundthat inequality declined among normal birth-weight births but widened substantiallyamong low birth-weight births. (They argued that the narrowing of inequality in infantmortality for normal birth-weight births occurred because affluent families reached theminim um achievable level of infant mortality.)A growing number of studies have examined inequalities in health and survival byhouseho ld economic status in less-developed co un tri es 5 Virtually all of them h ave fo-cused on infants and children, for whom the best data are available. These studies havedrawn on data sets with measures of household econom ic status, such as the World Bank'sLiving Standards Measurement Study (Wagstaff 2000; Wagstaff and Watanabe 2000), ordata sets with measures of household asset ownership and housing characteristics, suchas the Demographic and Health Surveys, together with techniques proposed by Filmerand Pritchett (1999, 2001) for converting these measures into a wealth index (Bonilla-Chacin and Ham mer 1999; Gwatkin et al. 2000). These new studies have presented con-centration curves and concentration indices-more-sophisticated measures of inequality

4. See, for example, Barros and Sawyer (1991), Carvalho (1974), Carvalho and Wood (1978), Castilla(1996), Daly (1985), Monteiro (1996), Monteiro et al. (1992), Sawyer, Fernandez-Castilla, and Monte-Mor(1987) , Schmer tmann and Sawyer (1996) , S imdes and Monte i ro (1995) , S imdes and Ol ive i ra (1997) ,Szwarcwald and C astilho (1995), and Wood and C arvalho (1988).5 The concept of pure inequalities in health-variation in health status across individu als in a popula-tion according to their health status itself-has recently been promoted for use by policy ma kers and researchers(see, e.g., Gakidou, Murray, and Frenk 2000; World Health Organiz ation 2000). However, there have been anumber of forceful critiques of this approach (see, e.g., Braveman, Starfield, and Geiger 2001; Szwarcwald2002). In this article, my primary interest is in so ioe onomi inequalities in health, which corresponds closelywith that of the vast majority of previous studies.


4/23

5ocioeconomic Inequalities in Mortalitythat have some important advantages over the rate ratios across groups that are typicallyused-to characterize levels of inequality in health by economic status. Few of these stud-ies have examined trends. They have, however, uncovered substantial cross-country dif-ferences in inequality in children's health and survival by household econom ic status. Ofconsiderable interest is that of the 9 countries examined by Wagstaff (2000) and the 44countries examined by Gwatkin et al. (2000), inequalities in under-five mortality byhousehold economic status were the highest for Brazil. Wagstaff (2000) found that theconcentration curve for Brazil lay the farthest from the 45-degree line and did not inter-sect with any o ther country's concentration cu rve, indicating that inequality in under-fivemortality by household econom ic status in Brazil w as unambiguously the worst.In this article, I examine the effects of social and economic development on socio-econom ic inequalities in under-five mortality for the state of SBo Paulo, Brazil, over a 21 -year period during which much of the transition in infant and child mortality unfolded. Iinvestigated whether the improv ements in infant and child survival were accom panied bydeclining inequalities by socioeconomic status or whether the most-advantaged segmentsof society benefited more from these im provemen ts. Focusing on inequality in under-fivemortality by household w ealth and mother's education, I drew on microdata from Brazil-ian censuses conducted in 1970, 1980, and 1991. The major strengths of these data in-clude the availability of detailed individual- and household-level covariates and extrem elylarge samples.I begin by providing some background on trends and differences in infant and childmortality in SBo Paulo over the past 3 0 4 0 years. In the subsequent section, I describe thedata and methods. I then present the results and end the article with so me conclusions.BACKGROUNDSBo Paulo is Brazil's largest state, with one fifth of the country's total population. Thestate, which is located in southeastern Brazil, had a population of 3 1.5 million in 1991(FundaqBo SEA DE 1993). The c ity of SBo Paulo and its surrounding m etropolitan area isthe most industrialized region in Latin America. This region, located in the southeasternpart of the state, forms Brazil's u rban-industrial heartland and dominates the country eco-nomically. For instance, the g reater SBo Paulo region generates on e third of Brazil's in-come, although it contains only one tenth of the country's population (Abranches 1995).During the past 30 years, the city and state have led Brazil and Latin America in manyimportant demographic and socioeconomic trends. For example, the state has the highestlevel of urbanization in the country, with 93 of the state's population living in urbanareas in the mid- 1990s.The decline in infant mortality in SBo Paulo state began in the early 1940s (see Fig-ure 1). It was interrupted by an increase in mortality between the mid-196 0s and the early1970s. From 1964 to 197 1, the infant mortality rate in SBo Paulo state rose by 26 , from7 1 per 1,000 to 89 per 1,000. How ever, it then d eclined rapidly, so that by 1980, it hadreturned to its long-term trend line. During the study period, the infant mortality ratedeclined from 84 per 1,000 in 1970 to 27 per 1,000 in 199 1.A number of studies have investigated why infant mortality rates increased in SBoPaulo betwee n 1964 and 197 1. One set of studies has argued that the economic policiesof the military governm ent that took power in 1964 were the main cau se of the upturn ininfant mortality (Sawy er 1981; Wood 1982; Yunes 1981). Althoug h there was rapideconom ic growth of over 10 per year beginning in 1964, there was a sharp concurrentdrop in real wages, with the legal minim um wage falling by approxim ately 60 between1964 and 1973. Other factors that may have contributed to the rise in infant mortalityrates include insufficient investment in water supply and sanitation (Monteiro andBenicio 1989) and in health services (Leser 1974) to keep pace with rapid rates of popu-lation growth.


5/23

446 Demography Volume 41-Number 3 August 2004Figure 1. Infant Mortality ate for SBo Paulo State: 1894 1998

Beg innin g in 1973, infant mortality rates began to decline in the state. Most research-ers have attrib uted the fall to improvem ents in infrastructure (Costa and D uarte 1989). Inthe 1970s, Brazil embarked on an intensive effort to expand the water-supply and sanita-tion system s (M errick 198 5), which raised the proportio n of households in Siio Paulowith running water from 71.3 in 1973 to 98.6 in 1984 (Sawyer et al. 1987). Otherimportant changes that contributed to lower mortality rates were longer durations ofbreast-feeding, the greater use of oral rehydration therapy, higher immunization rates,and an increase in the num ber of health cen ters (M onteiro et al. 198 9; Victora et al. 1996;Zuiiiga and Mo nteiro 1995).DATA AND METHO DSThis study was based on data from the survey component of the Brazilian populationcensuses that were conducted in 1970, 1 980, and 199 1. One quarter of the householdswere selected for the survey comp onent in 1970 and 1980; in 1991, 10 of the house-holds were selected in municipalities with a population of 15,000 or higher and 20 wereselected in the rest. Information was collected on the housing conditions and dem ographic,social, and econo mic characteristics of each resident. Most of the core qu estionnaire hadfew, if any, changes from census to census, which allowed me to create identical mea-sures for each year. The high sampling percentages resulted in exceptionally largesamples. Although m y analysis was based only on parous women aged 20-34, the samplesizes were 297,729 for 1970, 52 7,92 7 for 1980 , and 406 ,976 for 1991-a total of1,232,632 observations.I restricted the analysis to child mortality amo ng women aged 20-34 because amother's age was used to con trol for the duration of her children's exposure to the risk ofdeath. (Th e duration since the mother's first marriage is a commo nly used alternative to


6/23

7ocioeconomic Inequalities in Mortalitymother s age as a control for children s exposure; however, information on age at firstmarriage is not available from the Brazilian censu ses.) As a result, deaths to children ofolder women correspond to births occurring in the more distant past. Since the covariatesreflec t cond itions at the time of the interview, the inclusion of olde r women-and hen cechildren who were born in the more distant past-may have yielded misleading results,given the rapid changes in living conditions over the study period. Also, underreportingthe number of children born is more likely among older women. I excluded women aged15-19 because children of teenage mothers face higher mortality that is due , in part, totheir mothers age. (Although mother s age was measured and is included in the analysis,it is confounded with children s exposure.)The co re questionnaire collected information from w omen on the number of childrenborn an d the num ber of these children w ho were alive at the time of the interviews. I usedthis inform ation to construct an index o f child mortality for each mother using the tech-niques proposed by Trussell and Preston (1982 ). The index w as used to estim ate under-fivemortality proba bilities indirectly. Because the indirect estimates were based on individual-level index values, I was able to con struct comparab le indirect estimates of mortality thatcontrolled for individual- and household-level demographic, social, and economic charac-teristics using regression analysis. Next I describe the construction of the child mortalityindex (CMI), the regression analysis, and the measures of socioeconomic inequality inunder-five mortality.Index o Child MortalityThe CMI is based on the ratio of observed to expected deaths among all births to eachwom an in the sample aged 20-34 years. It assumes that a child s risk of death is propor-tional to a standard mortality schedule, q (a), and hence to the risk faced by other chil-dren. The probability of a child born to the jt h mothe r in the ith group dying by ex act agea , q,,(a), is given by

where p is the mothe r s proportionality factor. A model life table was chosen as the s tan-dard mortality schedule. The most important assump tions underlying the indirect estima-tion of child mortality are that the levels of fertility and childhood mortality have beenconstant in the recent past. Restricting the analysis to women ag ed 20-34 minimized theimpact of violations of these assumption^.^I exam ined deaths of children separately for subgroups of women on the basis of five-yea r age categories and levels of education because women of higher soc ioeconom ic statusgenerally bear children at a later age. Hence, among mothers of the same age, those ofhigher socioec onom ic status will have younge r children. These children are less likely tohave died because they were y ounger and because of their higher socioecon omic status. IfI ignored the fact that children of higher socioeconomic status are younger (holdingmothe r s age constant), the estimated mortality rates for these children would be under-stated . Consequ ently, it wa s important to control for fertility differences-and hence forchildren s e xpo sure to the risk of death-by mothers socio econ om ic status when I calcu-lated the CMI. I did so by stratifying mothers according to whether they had completedelementary school, an outcome that occurred well before the start of childbearing. This

6 Th e mean births to women aged 30-34 decreased from 3.1 in 1970 to 2.1 in 1991. Th e decline in fertilitywas larger amo ng less-educated w omen-with childbea ring at young er ages falling especially fast amon g thesewom en. For women aged 30-34 who had not complete d elementary sc hool, the mean parity fell by one third,from 4. 0 births in 1970 to 2.7 in 1991; for women with at least an eleme ntary school educa tion, fertility fell by209.6, from 2.5 births in 1970 to 2.0 in 1991.


7/23

448 Dem ography Volume 41-Numb er 3 August 2004procedure simply yields m ore-accurate estimates of the CM I and reduces any bias in theestimated relationship between socioeconomic status and under-five mortality.The expected numb er of dead children for the jt h woman in age group d and educa-tion group i E,, was given by

where B, is the num ber of children born to this wom an and PD ;(d) is the expected propor-tion o f children who died among wo men in five-year age group d and education gro up iunder the standard mortality schedule. I estimated PD:(d) by inv erting the convention alprocedure fo r estimating child mortality from sum mary information from mothers on thenum ber of children ever born and the numb er of children who d ied.The final step was to calculate the value of the CM I for each wom an. For the jthwom an in the ith education group, this index is the ratio of observed deaths (0 ,) to ex-pected deaths (E,,) of children :

CMI, 0 E,

On the basis of Eq. (I), the observed number of dead children can be written as P,E,, .Th us , the ratio of observe d to expected dead children-the CMI-is sim ply P,, the pro -portionality factor described earlier.I converted the CMI to an estimated under-five mortality probability, q(5), by multi-plying the q(5) value for the chosen standard life table by the weighted average of theindex-with the num ber of live births to each wom an serving as the weigh t (q(5) x 1,000is hencefo rth referred to as the under-jive m ortality rat e or simply un der-jive mortality). Ichose the Coale-Demeny West regional model life-table level 18.5 as the mortality stan-dard for this study on the basis of a preliminary analysis (results not shown) that usedindirect estimation techniques described in M an ua l X (United Nations 1983). Preston andHaines (1991) recommended using the q (5) life-table parameter to sum marize results be-cause it is likely to be the least sensitive to time trends or an error in the choice of amodel life table. The weighting procedure provided a useful way to com bine informationon child mortality across women of a wide age range (20-34 years) who had a corre-sponding large variation in the numb er of children born (and thus expo sed to the risk ofdeath). Values for the weighted average of q(5) are close to the corresponding un weightedestimates that are based only on the mortality of children to women aged 30-34, which isthe standard approach to estimating the q(5) parameter using indirect estimation tech-niques (United Nations 1983).Regression AnalysisI conducted a regression analysis of the CMI, which allowed me to estimate under-fivemortality rates after controlling for demographic, social, and economic characteristics.The regressions w ere run using weighted least squares with the child m ortality index asthe dependent variable and the number of live births to the mother as the weight. Thisapproach to regression analysis of the child mortality index w as proposed by Trussell andPreston (19 82) and has been applied elsewhere (e.g., Preston and Haines 1991); the re-sults were not sensitive to the choice of weighted least squares over an alternative, suchas Poisson regression.I used the regression m odels to produce adjusted estimates of under-five mortality bycalculating predicted values of the child mortality index while holding all variables con-stant at their samplewide means except the single independent variable of interest (wealth


8/23

9ocioeconomic Inequalities in Mortalityand, separately, mother s education), which retained its actual values. The cova riates in-cluded in the analysis were the household possessions-based wealth index (see the nextsection); household water supply; household sanitation; and mother s education, currentage, and interstate migration status.Measures of nequalityFinally, I used the under-five mortality estimates to calculate concentration curves andconcentration indices that describe and summarize socioeconomic inequalities in under-five mortality according to household econom ic status or mother s years o f education. Tomeasure t he family s eco nomic status, I used an index of household wealth that was basedon a principal-components analysis of the characteristics of housing and ownership ofconsumer durables (see Filmer and Pritchett 2001). The household-wealth index was thescore of the first principal com ponent and was based on the presence of electricity in thehousehold; ownership of a radio, refrigerator, television, and automob ile; and the numb erof bedrooms and bathroom s in the dwelling. I also calculated measures of socioeconom icinequality in under-five mortality according to the mothers years of education that re-flected the mothers highest level of educational attainment (rather than the numbe r ofyears o f schooling, which is affected by the repetition of grades).The concentration curve plots the cumulative proportion of mothers who wereranked in ascending order by household economic status or years of education (on the xaxis) against the cumulative proportion of under-five mortality (on the y-axis). If therewere perfect equality in children s deaths according to househ old econo mic status andmothers years of education, then the concentration curves would lie along the diagonal.The farther the concentration curves lie above the diagonal, the more that inequalities inunder-five mortality favor mothers from households of higher economic status or withmore education.Wh en one comp ares two concentration curves that do not cross, the one farther fromthe diagonal represents an unambiguously less-egalitarian distribution based o n any mea-sure of inequality that respects the principle of transfers (Atkinso n 1970). Wh en concen-tration curves cross, there are two ways to resolve the inherent ambiguity and to rankthem. The first is to transform them into generalized concentration curves (Shorrocks1983) by scaling them ac cording to the level of mortality in each population. Generalizedconcentration curves permit an absolute com parison of well-being and are much less likelyto cross than are relative concentration curves and hence ar e less likely to lead to ambigu-ous rankings. Th e second way to rank (relative) concentration curves that cross is to con-struct the corresponding concentration index for each curve (discussed next) and com parethe two values. The concentration index is one o f only two measures o f socioeconomicinequality in health to meet w hat Wagstaff, Paci, and van Doorslaer (19 91) argued are theminim al requirem ents for such a measure (the other is the slope index of inequality, whichis c lose ly related to the concentrat ion i n d e ~ ) . ~he standard concentration index reflects apartic ular chara cteriza tion of inequ ality aversion-specifically, on e that is sensiti ve tochanges in the m iddle of the distribution. Th is assumption is reflected in a set of weightsthat are assigned to the conc entration of ill health in different parts of the socioeco nomic-st atu s d i s t r i b ~ t io n . ~

7. The concentration curves and indexes did ot weight each mother-level observation by the number ofher children ever born because the inequality measures should not reflect differential fertility, which wouldhave been the case if I had used weighted values.8 . The requirements that Wagstaff et al. 1991) identified are 1) that the measure reflect the socioeco-nomic dimension of inequalities in health, 2) that it reflect the experiences of the entire population, and 3 ) thatit is sensitive to changes in the distribution of the population across socioeconomic groups.9. Wagstaff 2002b) developed an extended concentration index that can incorporate alternative patterns ofinequality aversion.


9/23

450 Demography Volume 41-Number 3 August 2004The concentration ind ex is defined as twice the area between the concen tration curveand the diagonal and is negative when the concentration curve lies above the diagonal.1The index is calculated as follows (see Kakw ani, Wagstaff, and van Doorslaer 1997):

2C = C x 4 -1,nx = I

where x is the under-five mortality index for the ith mother, R, (2i 1) 2n is therelative rank for the ith mother, and s the mean of the under-five mortality index. Thevariance for the concentration index (which takes into account serial correlation in thedata ) is calculated as

where

and

The standard concentration curve and index are well-suited for assessing changesover time in relative so cioeconom ic inequalities in under-five mortality, since these mea-sures are indepen dent of the absolute level o f mortality. Over the study period, how ever,there were large secular declines in under-five mortality rates. Generalized concentrationcurves (described earlier), which are scaled to the mean level of under-five mortality,provide a useful way to assess changes in socioeconomic inequalities in health that re-flect these improvements in survival.RESULTSI present m y results in two subsections. I begin by describing trends in levels of under-fivemortality and trends in inequality in under-five mortality by househo ld wealth and mothe r'seducation. Next, present adjusted measures of socioeconom ic inequality in mortality thatcontrol for the effects of a variety of demograp hic, social, and econom ic characteristics.Under-Five M ortality: Levels and InequalityThe under-five mortality rates fo r SBo Paulo s tate were 117.0 for 197 0, 96.0 for 1980, and45.7 for 1991 (see Table 1). Under-five mortality dropped by 61 over the entire 21-yearperiod. It fell by 18 between 1970 and 1980 and by 52 between 1980 and 199 1.The estimated levels of under-five mortality are consistent with infant m ortality ratesbased on vital statistics from FundaqBo SEADE, the SBo Paulo state statistical agency.The under-five mortality estimates based on data for wom en aged 20-34 refer, on aver-age, to approximately four years before the date of each census (see United Nations 1983for details on calculating the reference period for indirect estimates of child mortality).The infant mortality rates for 1966 (75.9 per 1,000), 1976 (77.2 per 1,000), and 1987(33.9 per 1,000) were all 76 -80 of the corresponding under-five mortality rate. M y

10. The conc entrat ion curve is the bivariate an alog of the Loren z curve, whereas the concentration index isthe b ivaria te analog of the Gini index. For the L orenz curve and G ini index, the sam e variable appears on boththe x-axis and the y-axis .


10/23

Socioeconom ic nequalities in MortalityTable 1. Levels and Inequality of Under-Five Mortality, by Area, for SBo Paulo State: 19 70,19 80 ,and 1991 (standard errors in parentheses)

Under-Five Concen t ra t ion Sample S izeArea Mor ta l i ty I nde x ( M othe r s )1970

Tota lU r b a nRural

1980Tota l

U r b a nRural

1991Total

U r b a nRural

Source Author s calculations based on microdata for S5o Paulo state from th e 197 0, 19 80 , and 1991 Brazilian censuses,

estimates and those based on vital statistics (see Figure 1) both suggest that it was notuntil the mid -1 970s that sharp an d sustain ed declines in infant and ch ild mortality o c-curred in SBo Paulo.The concentration curves fo r 1970, 1980, and 1991 indicate that inequality in under-five mortality by household economic status declined over the study period. Figure 2show s the concentration curves for 1970, 1980, and 1991, and Figure 3 show s the corre-sponding d eviations o f the concentration curves from the diagonal (which implies perfectequality). The tw o sets of curves are useful fo r examining inequality over the entire rangeof the econom ic-status measure, with the deviations curve providing the necessary detailto compare carefully the concentration curves for d ifferent years. I highlight two impor-tant findings. First, the 1980 curve do minates the curve for 1970; that is, the 1980 curveis below the 1970 curve over the entire distribution of wealth. This finding indicates thatthere was unam biguously less inequality in under-five mortality by household wealth in1980 than in 1970. Second, the 1980 and 199 1 curves overlap over much of the upper twothirds of the household-wealth d istribution, but the 1991 curve lies above the 1980 curvein the bottom third of the distribution. This finding indicates that there was a relativeincrease between 1980 and 199 1 in under-five mortality for wo men in the least-wealthythird of households. In fact, for the bottom fifth of households, there was a relative in-crease in under-five mortality compared with 1970. In contrast, there was little relativechange between 1980 and 1991 in under-five mortality for women in other parts of thehousehold-wealth distribution.The fact that the 1991 curve crosses both the 1970 and 1980 curves means that de-finitive and unambiguous rankings of inequality in under-five mortality by householdwealth cannot be m ade fo r comparisons involving 199 1. To address this issue, I examinedgeneralized con centration curves and the concentration index. The generalized concentra-tion curves for 1970, 1980 , and 1991 (presented in Figure 4) reflect the large decline inunder-five mortality that occurred between 1980 and 1991. In doing so, they provide away to resolve the ambiguity of the comparisons that are based o n the relative concentra-tion curves (see Figures 2 and 3). The results indicate that although there was no clear


11/23

452 Dem ography Volume 41-Number 3 August 2004Figure 2 Concentration Curves for Under-Five Mortality, by Househ old Wealth for Siio Paulo

State: 1970, 1980, and 1 991

0 ICumulative Proportion of Mothers,Ranked y Household W ealth

difference in relative inequality between 1980 and 1991, the large decline in under-fivemortality rates means that throughout the entire wealth distribution, under-five mortalitywas substantially lower in 1991 and hence that higher levels of well-being prevailed.Finally, in Table 1, I present estimates of the concentration index for 1970, 1980,and 1991 for the entire state of S5o Paulo and separately for rural and urban areas. Theconcentration index for the state rose substantially between 1970 and 1980, from -0.216to -0.163, indicating that there were lower levels of inequa lity in under-five mortality byhousehold wealth in the latter period. Subsequently, there was a small but statisticallysignificant decrease in the concentration index indicating higher levels of inequa lity inunder-five mortality by household wealth), with the concentration index reaching -0.175in 1991.Th e results presented in Table 1 suggest that the rise in inequality in und er-five mor-tality by household wealth between 1980 and 1991 for the state was due to a sharp in-crease in the index fo r rural areas since there were declines in the index for urban areas).Between 1980 and 1991 , inequality in under-five mortality by household wealth increaseddramatically in rural areas, to levels well abov e those found in 1970 Sastry 2004). Be-cause of these diverging trends for rural and urban areas, the subsequent analysis focusesexclusively on socioeconom ic inequalities in und er-five mortality for the urban areas ofS5o Paulo state.The concentration cu rves for urban areas of SBo Paulo state not shown ) revealedthat inequality in under-five mortality by ho usehold wealth was unamb iguously low er in


12/23

45ocioeconom ic lneaualities in Mortalitv

Figure3. ConcentrationCurveDeviationsforUnder-FiveMortality, by HouseholdWealthforSgoPauloState:1970,1980,and1991

1CumulativeProportionofMothers,RankedbyHouseholdWealth

both 1980 and 1991 than in 1970; however, the 1980 and 1991 curves intersect andhence cannot be rank ed definitively. T he generalized con centration curve fo r 1 99 1 dom i-nates (i.e., is everywhere lower than) the 1980 curve, indicating that although relativeinequality was similar for these two years, absolute levels of under-five mortality werefar lower in 1991 and hence that everyone was better off.The top panel of Table 2 presents summ ary information on differences and inequal-ity in under-five mortality by household wealth in urban areas of Siio Paulo state. Threemeasures of inequality are shown in Table 2: the ratio of under-five mortality in thelowest wealth quintile to that in the highest quintile, the difference in under-five mortal-ity between the lowest and highest wealth quintiles, and the concentration index. Allthree measures reveal a consistent decrease in inequality in under-five mortality byhousehold wealth between 1970 and 1991. The ratio of under-five mortality fo r the low-est to highest quintile of household wealth fell from 3.08 in 1970, to 2.72 in 1980, to2.22 in 1991. Over this period, the difference in under-five mortality between the firstand fifth wealth quintiles shrank relative to the overall level of mortality. n 1970, thedifference between the lowest and highest wealth quintiles was 116.4, while the overallunder-five mortality rate for urban areas wa s 114.4-indicating that the difference wasslightly greater than the overall mortality rate. By 1980, the difference of 91.4 wassmaller than the overall under-five mortality rate of 95.8, and by 1991, the difference of36.5 was only 80 of the overall mortality rate of 45.2. Finally, the concentration indexrose from -0.238 in 19 70, to -0.176 in 19 80, to -0.163 in 19 91. On the basis of the


13/23

454 Demography Volume 41-Number 3 August 2004Figure 4 Generalized Concen tration Curves for Under-Five Mortality by Household Wealth for

S lo Paulo State: 19 70 19 80 and 1991

0 1Cumulative Proportion of M othersRanked by Household Wealth

concentration index, inequality in under-five mortality by household wealth fell by onethird over the study period.An examination of und er-five m ortality rates by wealth qu intile reveals that the d e-cline in inequality in under-five mortality by ho usehold wealth was the result o f a muchlarger relative drop in under-five mortality for lower quintiles than for higher quintiles.Specifically, there was a 61 decline in under-five mortality for the lowest quintile be-tween 1970 and 1 991, which is one third larger than the 46 decline for the highestquintile over this period. Overall, and for each group, the magnitude of the declines inmortality was mu ch smaller for 1970-1980 than for 1980-1991. For the lowest wealthquintile, under-five mortality rates fell by 16 between 197 0 and 1980 but by 54 be-tween 1980 and 1991. For the highest quintile, under-five mortality fell by o nly 5 dur-ing the earlier period but by 44 during the later period.Turning to inequality in under-five mortality by mother's education, two main find-ings emerged from an examination of the concentration curves (not shown). First, in 1991,there was an unambiguo usly higher level o f inequality in under-five m ortality by m other'syears of education than in 1970 or 1980, since the 199 1 concentration curve lay above the1970 and 1980 curves everywhere. At the same time, the generalized con centration curvefor 1991 dominated the corresponding curves for 1970 and 1980, indicating that the de-cline in mortality meant that all mothers were better off in 1991 based o n improvem entsin children's survival. Secon d, the 1970 and 1980 concen tration curves intersected, al-though except for two small portions toward the bo ttom and top of the education distribu-tion, the 1980 curve is everywhere below the 1970 curve. However, because the decline


14/23

Socioeconomic lneaualities in Mortality 455Table 2. Levels and Inequality of Under-Five Mortality, by Wealth and Mother s Level of Education

for Urban Areas of Siio Paulo State: 1970,1980, and 1991 (standard errors in parentheses)1970 1980 1991

Percentage Percentage PercentageUnder-Five in Under-Five in Under-Five inVariab les Mor ta l i ty Category Mor ta l i ty Category Mor ta l i ty CategoryQuin t i le of Weal th

1 (lowest) 172.4 (1.04) 20.3 144.5 (0.84) 20.0 66.5 (0.58) 20.02 131.3 (0.96) 19.8 106.8 (0.73) 22.3 47.8 (0.51) 20.23 102.7 (0.87) 20.1 83.0 (0.71) 19.4 42.3 (0.47) 20.04 75.2 (0.79) 20.5 68.8 (0.66) 20.2 35.3 (0.44) 19.95 (highest) 56.0 (0.70) 19.3 53.1 (0.65) 18.1 30.0 (0.42) 19.8Total 114.4 (0.41) 100.0 95.8 (0.33) 100.0 45 .2 (0.22) 100.0Ratio: lowest to highest 3.08 1.05 2.72 1.10 2.22 1.01Lowest highest 116.4 91.4 36.5C oncen t r a t ion index -0.238 (0.002 7) -0.176 (0.002 5) -0.163 (0.0038)

Mother s Educat ionElem entary o r less

Illiterate 176.6 (1.18) 15.3 155.6 (1.20) 9.1 98.3 (1.35) 4.8Literate 104.0 (0.48) 69.7 98.5 (0.45) 55.9 55.6 (0.39) 36.7

M i d d l e 68.4 (1.22) 9.0 72.9 (0.70) 19.9 36.5 (0.35) 33.1Secondary 43.7 (1.35) 5.2 49.3 (0.89) 10.0 22.0 (0.39) 17.8Higher 29.7 (2.88) 0.8 36.9 (1.10) 5.1 14.9 (0.49) 7.6Total 114.4 (0.41) 100.0 95.8 (0.33) 100.0 45.2 (0.22) 100.0Ratio: lowest to highest 5.95 19.13 4.22 1.78 6.60 0.63Lowest highest 146.9 118.7 83.4C oncen t r a t ion index -0.208 (0.0028 ) -0.157 (0.0025 ) -0.274 (0.0037 )Nore The concentration index for mother s education is based on mother s years of educationource Author s calculation s based on microdara for Sio Paulo state from the 1970 , 1980, and 1991 Brazilian censuses.

in under-five mortality between 1970 and 1980 was relatively modest, even when it isfactored in (using the generalized concentration curves), the curves for 1970 and 1980cross. Thus no definitive ordering can be made regarding levels of inequality in under-five mortality by mother s education in 19 70 and 1980.The bottom panel of Table 2 presents summary information on differences and in-equality in under-five mortality by mother s level of education. I examined three sum-mary measures o f inequality in under-five mortality: the ratio o f under-five m ortality inthe lowest group (mothers with an elementary or lower level of education and who areilliterate) compared to the highest g roup (mothers with higher edu cation); the differencein under-five mortality between the lowest and highest educational groups; and the con-centration index (based on the mothers years of education). Although parallel to the mea-sures used for exam ining inequality in under-five mortality by household wealth, they aredifferent in one important aspect: the rate ratios and absolute differences are based onfixed-rather tha n relative-categories.


15/23

456 Demography Volume 41-Number 3 August 2004All three summary measures indicate that inequali ty in under-five mortali ty bymoth er's education fell between 1970 and 1980 but then increased in 199 1 to a levelabove that in 1970. For example, the ratio of under-five mortality for illiterate motherscompared to mothers with higher edu cation was 5.95 in 1970, 4.22 in 1980, and 6.60 in

1991. There was a large improvement in average educational levels over this period, asreflected in the distribution of mothers in the sample across educational categories. In1970, 15.3 of the mothers were illiterate, and only 0.8 had a higher education; by1991, however, fewer than 5 of the mothers were illiterate, and 7.6 had a higher edu-cation. This absolute improvement in education could affect the assessment of changes ininequality in under-five mortality by mother's education, especially when these resultsare compared to those based on inequality in wealth. In particular, increases in the ratioof mortality among the least-educated group to the m ost-educated group may be less im-portant because the number o f women in the former group declined while the number inthe latter group increased. However, the results for the concentration index, which is basedon mo thers' single years of education, suggests that this was not the case because exactlythe same findings are apparent when this measure is examined. The con centration indexincreased from -0.208 in 19 70 to -0.157 in 1980, but then dropp ed to -0.274 in 19 91.Over the full study period, the level of inequality in under-five mortality by mother'seducation, based on the concentration index, increased by on e third.On e interesting result with regard to unde r-five mortality by moth er's level of educa-tion is that mortality incre sed between 1970 and 1980 for wo men with middle, second-ary, or higher education. Only for women w ith elementary or less education was there adecrease in under-five mortality over this period. It is unclear what accounted for thisresult, although it may have been caused, in part, by the large increases in educationalattainment between 1970 and 1980 that shifted women from disadvantaged backgroundsup the educational d istribution. Between 1980 and 199 1, however, all groups exp eriencedsubstantial declines in und er-five mortality. For the low est group (illiterate wom en), therewas a 37 decline in under-five mortality; for the highest group (women with highereducation), there was a 60 decline.Adjusted Levels of Socioeconom ic Inequality in Under-Five MortalityThe results presented so far have shown raw differences and inequalities in under-fivemortality rates by socioeconomic status. Of considerable interest is the extent to whichunderlying differences in demographic, social, and economic characteristics shaped theseresults. For instance, low levels of under-five mortality among the m ost-educated womenmay reflect, in part, the survival advantages conferred by higher household wealth. Bycontrolling for the set of covariates iden tified earlier, I was able to examine inequality inunder-five mortality by mother's level of education net o f other key factors.I begin with adjusted inequalities in under-five mortality by household wealth acrossall urban areas. The results suggest that after I controlled for demographic, social, andeconomic factors, there continued to be a decline in inequality in under-five mortality byhousehold wealth between 1970 and 1980, but little change in inequality between 1980and 199 1. The concentration curves (not shown) cou ld be o rdered clearly, with the 1980curve dominating the 1970 curve and the 1991 curve lying entirely between the curvesfor 1970 and 1980. The concentration index had the same trend (see the top panel ofTable 3), rising from -0.138 to -0.070 between 1970 and 1980 and falling to -0.078 in199 1 The chang e in the con centration index between 1980 and 199 1 was no t statisticallysignificant and, indeed, the concentration curves for these two years lie close to eachother. The results based on the rate ratio and the absolute difference between the lowestand highest wealth qu intiles were consistent with those based on the concentration curveand index for the change between 1970 and 1980, but they were discrepant for the changebetween 1980 and 1991 (which was, however, small and unimportant). The results based


16/23

57ocioeconom ic Inequa lities in MortalityTable 3. Levels and Inequality of Adjusted Under-Five Mortality, by Wealth and Mother s Levelof Education for Urban Areas of S b Paulo State: 19 70 ,1 98 0, and 1991 (standard errors

in parentheses)1970 1980 1991Percentage Percentage PercentageUnder-Five in Under-Five in Under-Five in

Variables Mortality Category Mortaliry Category Mortality CategoryQuin tile of Wealth

1 (lowest) 1 3 6 1 (0.80) 20.32 119.9 (0.52) 19.83 106.6 (0.42) 20.1

5 (highest) 70.5 (0.90) 19.3 71.7 (0.66) 18.1 32.7 (0.40) 19.8Total 114.4 (0.41) 100. 0 95.8 (0.33 ) 100.0 45.2 (0.22 ) 100.0Ratio: lowest to highest 1.9 3 1.05 1.55 1.10 1.51 1.01Lowest highest 65.6 39.7 16.8Concentration index -0.138 (0.00 26) -0.070 (0.0025) -0.078 (0.00 37)

Mother's EducationElem entary or less

Illiterate 141.2 (0.99) 15.3 131.5 (1.00) 9.1 86.1 (0.87) 4.8Literate 100.1 (0.50) 69.7 95.5 (0.45) 55.9 50.9 (0.35) 36.7Middle 93.9 (1.58) 9.0 77.4 (0.81) 19.9 36.9 (0.39) 33.1

Secondary 87.2 (2.22) 5.2 69.2 (1.23) 10.0 25.4 (0.57) 17.8Higher 79.0 (5.56) 0.8 67.5 (1.74) 5.1 19.8 (0.91) 7 .6Total 114.4 (0.41) 100.0 95.8 (0.33) 100.0 45.2 (0.22) 100.0Ratio: lowest to highest 1.79 19.13 1.95 1.78 4.35 0.63Lowest highest 62.2 64.0 66.3Concentration index -0.068 (0.0028) -0.086 (0.0024) -0.194 (0.0036)Notes: Adjusted estimates control for covariates and set each covariate to its mean across all urban areas. The concentrationindex for mother's education is based on mother's years of education.Source: Author's calculations based on microdata for Si o Paulo state from the 1970 , 19 80, and 1991 Brazilian censuses.

on the wealth quintiles, presented in Table 3 suggested that there was a consistent declinein inequality in under-five m ortality by househo ld wealth. The ratio of unde r-five mortal-ity for the lowest quin tile to the highest decreased from 1.93 in 1 970, to 1.55 in 1 980, to1.5 1 in 1991 . The absolute difference between the lowest and highest quintiles narrowedover this period, from 65.6 in 1970 (57 of the overall level), to 39.7 in 1980 (41 of theoverall level), and to 16.8 in 1991 (37 of the overall level).For each year, there was substan tially less inequ ality in under-five mortality by h ouse-hold wealth based o n adjusted under-five mo rtality than on observ ed under-five mortality.For example, the adjusted concentration index was 42 lower than the unadjusted index in1970, 60 lower in 1980, and 52 lower in 1991. The independent relationship betweenhousehold w ealth and under-five mortality thus accounted for roughly half the observedinequality in u nder-five mortality by h ousehold w ealth, while the other covariates included


17/23

458 Demography Volume 41-Number 3 August 2004in the models accounted for the remainder. That is, the difference between th e adjusted andunadjusted levels of inequality in under-five mortality by household w ealth was the resultof other covariates-including household water supply and sanitat ion and mother 'seducation-that were positively correlated with house hold wealth and negatively relatedto the risk o f under-five m ortality.Over the full study period, there w as a more m arked drop in inequality in under-fivemortality by household wealth after other demographic, social, and economic characteris-tics were controlled ( 4 3 for the concentration index) than before (-32 for the concen-tration index ). Thus, chang es between 1970 and 199 1 in the levels and effects of covariatesthat were used to calculate adjusted mo rtality levels together serv ed to redu e the observeddecline in inequality in under-five mortality by household wealth. In other words, theobserved decline in inequality in under-five mortality by household wealth between 1970and 1991 was smaller than it otherwise would have been because the model covariatestogether operated in the opposite direction to widen these inequalities. In particular,moth er's education-which was positively associated with househ old wealth-had cova-riate effects that grew larger and stronger over the study period, although inequality inmoth er's education itself actually declined. This change in the effect of moth er's educationreduced the observed decline in inequality in under-five mortality by household wealtheven as other covariates, such as household sanitation and water supply, had a compensat-ing effect by accounting for part of the mortality advantage that wealthier householdsenjoyed in earlier years. From a policy perspective, these results highlight the importanceof considering the equity effects of other policies, programs, and secular changes when thegoal is to reduce economic inequality in health. So me factors, such as improvem ents ininfrastructure, provide alternative pathways through w hich to reduce economic inequali-ties in health instead of simply reducing household-wealth disparities themselves. How-ever, the effect of moth er's education in this case work ed in the opposite direction toincrease inequality in und er-five m ortality by household w ealth.By adjusting for covariates, I found that a clearer picture emerged of trends in under-five mortality inequality by mother's level of education. A comparison of the adjustedconcen tration curves indicated that there was unam biguo usly higher-and substantiallymore-inequality in under-five mortality by moth er's edu catio n in 1991 than in the twbearlier periods. On the other hand. the 1970 and 1980 concentration curves crossed whenI considered both the relative and the generalized curves. The summary measures, pre-sented in the bottom panel of Table 3 also showed that adjusted inequality in under-fivemortality by mother's level of education increased dramatically over the entire studyperiod. Th e ratio o f under-five mo rtality for the least-educated gro up to that for the most-educated group rose from 1.79 in 1970, to 1.95 in 1980, to 4.35 in 1991. Between 1970and 1991, the absolute gap in under-five mortality between the least-educated and themost-educated groups increased slightly (by 7 ), even though the overall level of under-five mortality declined substantially (by 60 ). Finally, the concentration index, whichwas based on mother's years of education , declined greatly over time, from -0.068 in1970 , to -0.086 in 1980, to -0.194 in 1 991, indicating a large increase in inequality inunder-five mortality by mother's education.The cov ariates-in ;he mod els accounted fo r a substantial proportion o f the inequa lityin under-five m ortality by mother's education in 1970 but a considerably sm aller propor-tion in 198 0 and 199 1. In particular, the covariates accounted for 67 of the unadju stedconcentration index in 1970 , 45 in 1980, and 29 in 1991. This finding suggests thatdemographic factors, living conditions, and household wealth played a significantly lessimpo rtant role over time in accounting for the observed ineq uality in under-five mortalityby mother's education. Instead, the independent effects of maternal education emergedover time as by far the most powerful factor behind th e observed inequality in under-fivemortality by mother's education. That is, observed differences in under-five mortality by


18/23

59ocioeconomic Inequa lities in Mortalitymoth er's education were increasingly due to mother's education itself, rather than to oth ercharacteristics-such as househ old wealth-that were correlated with moth er's education .The adjusted decline in under-five mortality was small for women at the bottom of theeducational distribution and large for women at the top. Adjusted under-five mortalitydeclined only 40 for illiterate women but 75 for wo men with higher education. Thedecline for women with higher education was nearly twice as large as the decline forilliterate wom en and 50 higher than the unadjusted decline for these same wom en. Notonly were wo men's higher levels of education more beneficial for their children's survival,but there was also a substantial redistribution of wo men up the edu cation distribution.There was an almo st twofold increase between 1970 and 1991 in the adjusted con-centration index for under-five mortality by mother's education, compared with a 32increase in the unadjusted concentration index. Hence, the total effect of the covariatesthat were controlled for in the model was to hold the observed increase in inequality inunder-five mortality by mother's education to a much smaller magnitude than it other-wise m ight h ave been. In other wo rds, the total effect of all other covariates-including,for instance, househ old wealth-was to reduce inequalities in unde r-five mortality bymother's education. This finding has some important policy implications. In particular,on the basis of past trends, a reduction in inequality in under-five mortality by mother'seducation is unlikely to be achieved by further increasing average educational levels orby reduc ing ineq uality in education. To lower the level of inequality in unde r-five mortal-ity by mother's education, there is a particular need to reduce mortality amon g the chil-dren of mothers with the lowest levels of schooling. Policies and programs to achieve thisgoal shou ld focu s on either directly targeting child-surviv al interventions toward the least-educated mothers o r improving other key factors, such as household wealth or the watersupply and sanitation, that appeared to have particularly large benefits for the chances ofsurvival of children in disadvantaged households.CONCLUSIONSIn this article, I presented trends in socioeconomic inequality in under-five mortality forSZo Paulo state from 197 0 to 199 1. Durin g this period , there were m ajor d eclines in under-five mortality for the state that were associated with substantial improvements in livingconditions, increases in educational attainment, and other demographic and socioeconomicchang es. Inequality in under-five mortality by household wealth clearly declined over thisperiod, with the drop concentrated in the 1970-1980 period. Inequa lity in under-five mor-tality by mo ther's education first declined and then increased, with a net rise over the studyperiod. When I controlled for background dem ographic, social, and econom ic characteris-tics, inequality in under-five mortality by household wealth fell even more than didobserved inequality between 1970 and 1980 (but remained constant thereafter). On theother hand, adjusted inequality in under-five mortality by mother's education increasedsubstantially, both between 1970 and 1980 and between 1980 and 1991, but especiallyduring th e latter period.The results suggest that , on the one hand, changes in the relationship betweenmother's education and the risks of under-five mortality increased socioeconomic in-equalities in under-five mortality by both household wealth and mother's education inSgo Paulo between 1970 and 1991. On the other hand, changes in the effects of householdwealth on under-five mortality reduced these inequalities. Improvements in householdwater supply and sanitation and changes in migration patterns had similar effects on so-cioeconomic inequalities in under-five mortality to those of household wealth. It appearsthat the poorest households benefited the most from these improvements. Thus, socialand econom ic development (other than improvements in mother's education) were asso-ciated with low er levels of socioecon omic inequality in mortality, sug gesting that the dis-advantaged did indeed benefit from progress over this period. In addition, these changes


19/23

460 Demography Volume 41-Number 3 August 2004kept the increase in inequality in under-five mortality by mother's education between1970 and 1991 smaller than it otherwise might have been.The preceding d iscussion focused on rel tive socioeconomic inequality in und er-fivemortality-that is, inequality independ ent of the level of mortality. Ove r the study period,under-five mortality declined substantially-in large part, because of impro vemen ts inmo ther 's education-so that althoug h relative inequality increased, all socioecon om icgroups enjoyed lower under-five m ortality rates at the end o f the study period than at thebeginning. N evertheless, relative socioeconomic inequalities in health are im portant frommany different perspectives. For example, these inequalities are generally perceived asunjust and unfair by policy makers and the public and are the target of programs andpolicies to reduce them.Comp aring the adjusted and unadjusted concentration indexes for under-five mortal-ity based on mother's education with the corresponding indices based on householdwealth suggests that in 1991, mother's education was roughly twice as important in ac-counting for socioeconomic inequality in under-five mortality. Thus, mother's educationappears to have emerged as probably the key factor underlying socioeconomic inequali-ties in under-five mortality-either through selection effects or through causal effects,whereby schooling provides women with the knowledge, means, and ability to raisehealthy children (Caldwell 19 79). There were enormous improvements in wom en's edu-cation over the study period. The average number o f years of education for women in thesample increased from 3.1 in 1970 to 5.1 in 1980 to 6.4 in 1991. This increase wasassociated w ith a large decline in the number o f women with the lowest levels of educa-tion, but was also remarkably associated with steady declines in inequality in schooling(Lam and Duryea 1999). The Gini index for years of schooling among women in urbanareas of SZo Paulo fell from 0.417 in 1970 to 0.391 in 1 980 to 0.329 in 1991, withinequality in women's schooling being unamb iguously lower in 1991 than in 1970 or1980, according to both the standard and generalized concentration curves. Note thatthere has been little indication that inequalities in pure income or wealth have been de-clining in Brazil. For example, a recent study found that income inequality remainedroughly constant for urban areas of the country between the 1970s and 1990s (Ferreiraand Paes de Barros 1999).A topic of current research and policy interest is the relationship between economicgrowth and inequalities in health by economic status. Contoyannis and Forster (1999)showed that no predictions could be m ade abou t the effects of income growth on healthinequality by econom ic status. Wagstaff (2002 a) sugg ested, how ever, that evidence-fromboth cross-sectional and trend analyses-is consistent with the association of higher aver-age incomes with higher levels of inequality in health by economic status. Victora et al.(200 0) conte nded that technological change-which generally accompanies incomegrowth-contributes to widen ing health inequalities by econom ic status because the moreadvantaged tend to benefit sooner from new medical know ledge and treatments. Finally,Szreter (1997), drawing on h istorical evidence from B ritain, argued that rapid economicgrowth results in worse population health, such as increasing inequalities in health, be-cause economic growth is associated with widespread and pervasive disruption. Unlessthis disruption is mediated by effective social and political responses, disease and deathwill result.

However, my results, which provide precise estimates of trends in inequality inhealth by economic status over a far longer period than did those presented in Wagstaff(20 02 a) or Victora et al. (2000 ), do not sup po rt these conclusions-at least for SZo

11. Szreter's (19 97) characterization may certainly he lp explain SBo Paulo's period of rising infant andchild mortality between the mid-1 960s and the mid-1970s.


20/23

Socioeconom ic Inequalities in Mortality 46 Paulo. In particular, the average annual growth rate in the per capita gross domesticproduct for Brazil was 8.5% between 1970 and 1980 and 2.2% between 1980 and 1991(Easterly and Sewadeh 2002). Growth was steady over the first period, but there wereseveral years with negative growth over the second period. However, observed inequal-ity in under-five mortality by household wealth for SBo Paulo declined both between1970 and 1980 and between 1980 and 1991. In fact, the largest decrease in inequ ality inunder-five mortality by household w ealth occurred between 1970 and 1980. (In contrast,the largest decrease in under-five mortality levels occurred between 1980 and 1991.)Thus, rather than being closely tied to economic growth, as suggested by the previousliterature, the trends in inequality in under-five mortality by household wealth in SBoPaulo were influenced in more complex ways by improvements in sanitation and thewater supply, rising levels of education for wom en, and a catching-up effect followingthe increase in mortality in the late 1960s and early 1970s. Finally, my results suggestthat higher levels of women's education were associated with higher levels of socioeco-nomic inequality in health.All cross-country studies of inequality in under-five mortality that were reviewedearlier have drawn on survey data, which have a numb er of important strengths-for ex-ample, the samples and survey measures tend to be highly comparable across countries,and the surveys collect information on children's health status in addition to measures ofchild mortality. However, they have some important weaknesses compared with censusdata. In particular, using census data and indirect estim ation techniques provides an alter-native source of information on under-five mortality that has the advantage of substan-tially larger sample sizes that allow considerably more statistical precision in trackingchanges over time. As I have shown, census data also allow researchers to uncover his-torical trends in socioeconom ic ineq ualities in mortality.REFERENCESAbranches, S. 1995. Crise e M udanqa [Crisis and change]. G eneva: International Labor Office.Atkinson, A.B. 19 70. On the Measurement of Inequality. Journal o fEconomic Theory 2:244-63.Barros, R.P. and D.O. Sawyer. 199 1. Unequal Opportunities to Survive: Education and Regional

Disp arities in Brazil. Working paper. Institute de Pesquisa E conbmica Aplicado, Rio de Janeiro.Bonilla-Chacin, M. and J. Ham mer. 1999. Life and Death Am ong the Poorest. Working paper.

World Bank, Washington, DC.Braveman, P., B. Starfield, and H.J. Geiger. 200 1. World Health Report 2000: Ho w It Remov es

Equity From the Agenda for Public Health Monitoring and Policy. Brit ish Medical Journal323~678-81.

Braveman, P. and E. Tarimo. 2002. Social Inequalities in Health Within Countries: Not Only anIssue for Affluent Nations. Social Science and M edicine 54: 1621-35.

Caldwell, J.C. 1979 . Education as a Factor in Mortality Decline: A n Examination of NigerianData. Population Studies 33:395-413.

Carvalho, J.A .M . 1974. Regional Trends in Fertility and Mortality in Brazil. Population Studies28:401-21.

Carvalho, J.A.M. and C.H . Wood. 1978 . Mortality, Inco me Distribution, and Rural-Urban Resi-dence in Brazil. Populat ion and Developm ent Review 4:405-20.

Castilla, F.M. 1996. Uma An alise Regional dos Differenciais Socio-EconB micos da Mortalidadena Infsncia no Brasil, 1960-80 [A Regional analysis of socioeconomic differentials in infantmo rtality in Braz il, 1960-801. Pp. 2235-52 in Anais do Encontro Nacional de EstudosPopulacionais Belo Horizonte, B razil: AssociaqBo Brasileira de Es tudos Populacionais.

Cleland, J. , G. Bicego, and G . Fegan. 199 2. Socioeconom ic Inequalities in Childhood Mortality:The 1970s to the 1980s. Health Transition Rev iew 2:l-18.Contoyannis, P. and M. Forster. 1999. The Distribution of Health and Income: A TheoreticalFramework. Journal of Health Econo mics 18:605-22.


21/23

462 Demography Volume 41-Number 3 August 2004Costa, N.R. and C.M .R. Duarte. 1989. "Notas Para AvaliaqBo de Politicas Sociais: A Te ndincia da

Mortalidade Infantil nas ~ l t i m a s Ccadas" [Notes for an evaluation of social policies: The trendof infant mortality in the last decades]. Dados 32:241-55.

Daly, H . E . 1985. "Marx and Malthus in Northeast Brazil: A Note on the World's Largest ClassDifference in Fertility and Its Recent Trends." Population Studies 39:329-38.DaVanzo, J. and J-P. Habich t. 1986. "Infant M ortality Decline in Malaysia, 1946-1975: The Roles ofChange s in Variables and Changes in the Structure of Relationships." Demography 23: 143-60.

Easterly, W. and M. Se wa deh . 2002. Global Develo pmen t Netw ork Growth Database. Washington,DC : World Bank. Available on-line at http://www.worldbank.org/research/growth/GDNdata.htmFerriera, F.H.G. and R. Paes de Barros. 1999. "The Slippery Slope: Explaining the Increase inExtreme Poverty in Urban Brazil, 1976-1996." Brazilian Review of Econometrics 19:211-96.Filmer, D. and L. Pritchett. 1999. "The Effect of Household Wealth on Educational Attainment:

Evidence From 35 Countries." Population and Development Review 25:85-120.2001. "Estimating Wealth Effects Without Expen diture Data-or Tears: A n Application to

Educational Enrollments in States of India." Demography 38 : 115-32.FundaqBo SEADE. 1993. 0 Novo Retrato de SGo Paulo: Avaliaqdo dos Primeiros Resultados doCens o Demogra co de 1991 [The new picture of SBo Paulo: Evaluation of the first results ofthe dem ographic census o f 19911. 2nd e d. SZo Paulo: FundaqBo SEAD E.

Gakidou, E .E., C.J.L. Murray, and J. Frenk. 2000. "Defining and Measuring H ealth Inequality: AnApproach Based on the Distribution of Health Expectancy." Bulletin ofthe World Health Orga-nization 78:86-96.

Gwatkin, D .R. 2000. "Health Inequalities and the Health of the Poor: What Do We Know ? WhatCan We Do?" Bulletin o f the World Health Organization 78:3-17.Gwatkin, D.R., S. Rutstein, K. Johnson, R.P. Pande, and A. Wagstaff. 2000. "Socioeconomic Dif-ferentials in Health, Nutrition, and Population." Working paper. World Bank, Washington, DC.

Kakw ani, N., A. Wagstaff, and E. van Doorslaer. 1997. "Socioec onomic Inequalities in Health:Measurement, Computation, and Statistical Inference." Journal of Econometrics 77:87-103.Lam , D. and S. Duryea . 1999. "Effects o f Schooling on Fertility, Lab or Supply, and Investme nts inChildren, With Evidence From Brazil." Journal of Human Resources 34: 160-92.

Lam, D. and D. Levison. 1991. "Declining Inequality in Schooling in Brazil and Its Effects onInequality in Earnings." Journal o f Development Economics 37:199-225.

Leser, W. 1974. "Crescimento da PopulaqBo e Nivel de Saude na C idade de SBo Paulo" [Populationgrowth and the level of health in the city of SBo Paulo]. Problemas Brasileiros 16:17-36.

Merrick, T.W. 1985. "The Effect of Piped Water on Early Childhood Mortality in Urban Brazil,1970 to 1976." Demography 22:l-23.

Monteiro, C.A. 1996. 0 Panorama da NutriqBo Infantil Nos Anos 90" [The panorama of infantnutrition in the 19 90 ~1 . iscussion document No. 1. UNICEF, Brasilia, Brazil.Monteiro, C.A. and M.H .A. Benicio. 1989. "Determinants o f Infant Mortality Trends in Develop-ing Cou ntries: So me Evidence From SBo Paulo City." Transactions of th e Royal Society of Tropi-cal Medicine and Hygiene 83:s-9.Monteiro, C.A ., M.H.A. Benicio, R. Iunes, N.C. G ouveia, J.A.A.C. Taddei, and M.A.A. Cardoso.1992. "Nutritional Status of Brazilian Children: Trends From 1975 to 1989." Bulletin of theWorld Health Organization 70:657-66.Mon teiro, C.A ., H.P.P. Zuniga, M .H.A. Benicio , and C.G. Victora. 1989. "Better Prosp ects for Child

Survival." World Health F orum 10:222-27.Nguyen, N.N. and A Wagstaff. 2002. "Poverty and Survival Prospects of Vietnamese ChildrenUnder Doi Moi." Policy Research Working Paper No. 2832. World Bank, Washington, DC.Preston, S.H. and M.R. Haines. 1991. Fatal Years: Child Mortality in Late Nineteenth-CenturyAmerica. Princeton, NJ: Princeton University Press.Psacharopoulos, G., S. Morley, A. Fishbone, H. Lee, and W.C. Wood. 1995. "Poverty and IncomeInequality in Latin America During the 1980s." Review o f lnc om e and Weal th 41:245-64.
http://www.worldbank.org/research/growth/GDNdata.htmhttp://www.worldbank.org/research/growth/GDNdata.htm


22/23

46ocioeconomic Inequalities in MortalitySastry, N. 1996. Comm unity Char acteristics, Individual and Household Attributes, and Child Sur-

vival in Brazil. Demography 33:211-29.2004. Urbanization, Development, and Under-Five Mortality Differentials by Place of

Residence in S l o Paulo, Brazil, 1970-1991 . Demo graphic Research (Special Collection 2, Ar-ticle 14). Available on -line at http://www.demographic-research.org/special/2/14/s2-14.pdfSawyer, D.O. 1981. Effects of Industrialization and Urbanization on Mortality in DevelopingCountries: The Case of Brazil. International Population Conference Manila 1981. Liege, Bel-

gium: International Union for the Scientific Study of Population.Sawyer, D.O., R. Fernandez-Castilla, and R.M . Monte-Mor. 1987. The Impact of Urbanizationand Industrialization on Mortality in Brazil. World Health Statistics Quarterly 40:84-95.

Schm ertmann , C.P. and D.O. Sawyer. 1996. Migration B ias in Indirect Estimate s of Regiona lChildhood Mortality Levels. Mathematical Population Studies 6:69-93.Shorrock s, A.F. 1983. Ranking Incom e Distributions. Economica 50:3-17.

Simdes , C.C.S. and C.A. Monteiro . 1995. Tendencia Secular e Diferenciais Regionais daMor talidade Infantile no Brasil [Secular trends and regional differentials in infant mortality inBra zil]. Pp. 153-56 in Velhos e Novos M ales da Saude no Brasil edited by C .A. Monteiro. SBoPaulo: H ucitec.Simdes , C.C. and L.A.P. Oliveira. 1997. A Saude Infantil no Brasil No s Anos 90 [Infant health inBrazi l in the 1990~1.Working paper. Instituto Brasilero de Geografia e Estatistica, Rio deJaneiro.

Stecklov, G., A. Bommier , and T. Boerm a. 1999. Trends in Equity in Child Surviv al in Develop -ing Countries: An Illustrative Analysis Using Uganda n Data. Working paper. Carolina Popula-tion Center, University of North Carolina, Chapel Hill, NC.

Szwarcwald, C.L. 2002. On the World Health Organization's Measurement of Health Inequali-ties. Journal of Epidemiology and Comm unity Health 56: 177-82.Szwarcwald, C.L. and E.A. Castilho. 1995. Estimativas da Mortalidade Infantil no Brasil, Decadade Oitenta [Estimates of infant mortality in Brazil in the 1 98 0~ 1. evista de Satide Pziblica29:45 1-62.

Szreter, S. 1997. Economic Growth, Disruption, Deprivation, Disease, and Death: On the Impor-tance of the Politics of Public Health for D evelopment. Population and Development Review23:693-728.

Trussell, J. and S. Preston. 1982. Estimating the Covariates of Childhood Mortality From Retro-spective Reports of M others. Health Policy and Education 3 :1-36.United Nations. 1983. Indirect Techniques for D emo graph ic Estimation Ma nual X. New York:

United N ations.Victora, C.G ., M.T. Olinto, F.C. Barros, and L.C. Nobre . 1996. Falling Diarrhoe a Mortality inNortheast Brazil: Did ORT Play a Role? Health Policy and Planning 11132-4 1.

Victora, C.G ., J.P. Vaughan, F.C. Barros, A.C. S ilva, and E. Tomasi. 2000 . Explaining Trends inInequities: Evidence From Brazilian Health Studies. Lancet 356: 1093-98.Wagstaff, A. 2000. Socioec onomic Inequalities in Child Mortality: Comparisons Across Nine De-

veloping Countries. Bulletin of the World Health Organization 78: 19-29.2002a. Inequalities in Health in Developing Countries: Swimming Against the Tide?

Policy Research Working Paper 2795. World Bank, Washington, DC.. 2002b. Inequality Aversion, Health Inequalities, and Health Achiev ement. Journal of

Health Economics 2 1 6 27-41.Wagstaff, A. P. Paci, and E. van Doorslaer. 1991. On the Me asurem ent of Inequalities in Health.Social Science and Medicine 33:545-57.

Wagstaff, A. and N . Watanabe. 2000. Socioeconomic Inequalities in Child Malnutrition in theDeve loping World. Policy Researc h Working Paper No. 2434. World Bank, Washington DC.Wood, C.H. 1982. The Political Economy of Infant Mortality in S l o Paulo, Brazil. InternationalJournal of Health Services 12:2 15-29.
http://www.demographic-research.org/special/2/14/s2-14.pdfhttp://www.demographic-research.org/special/2/14/s2-14.pdf


23/23

464 Demography Volume 41-Number 3 August 2004Wood, C.H. and J.A.M. Carvalh o. 1988. The Demography ofln equa lity in Brazil. New York: Cam -

bridge University Press.World Health O rganization (W HO). 2000. World Health R eport 2000. Geneva: WHO.Yunes, J . 198 1. Evolution of Infant Mortality and Proportional I nfant Mortality in Brazil. WHO

Statistical Quarterly 34:200-19.Zuiiiga, H.P.P. and C.A . Mo nteiro. 1995. Uma Nova Hip6 tese Para a AscensZo da Mor talidadeInfantil da Cidade de SZo Paulo Nos Anos 60 [A new hypothe sis for the rise in infant mortalityfor the city of SZo Paulo in the 19 60 ~1 . p. 157-72 in Velhos e Novos Males da Saude no Brasiledited by C .A . Montiero. SZo Paulo: Hucitec.

mortalidad infantil e inequidades socioeconómicas brasil

Documents